Momentum is building among integrated circuit vendors to change the chip industry. There are global efforts to make the semiconductor supply chain more sustainable. Leading manufacturers are openly committing to deadlines for carbon neutrality. It will take time, money and effort, as we discuss in this issue’s cover story.
Sustainability is big business. There is a globally accepted imperative to reduce the impact industrialization is having on the environment. Efforts have accelerated over the last decade, as more studies warn us of what lies ahead.
As a leading supplier of electronic components, Avnet takes its environmental responsibilities seriously. Avnet is an active member of the Electronic Components Industry Association (ECIA) and the Semiconductor Industry Association (SIA). With these and other organizations, Avnet is helping to address the global electronics industry’s environmental impact. As a distributor, Avnet has limited influence on the semiconductor manufacturing processes used but acknowledges their importance. Here, we provide an overview of the efforts to improve the environmental credentials of this sector.
The chip industry, fundamentally companies supplying process equipment, recognizes that semiconductor manufacturing consumes large amounts of resources and produces significant amounts of greenhouse gasses (GHGs). A large semiconductor fab can use the same amount of power as 50,000 homes and as much as 9 million gallons of water per day.
In November 2022, SEMI, the electronics manufacturing and supply chain industry association, launched the Semiconductor Climate Consortium (SCC) with 60 founding members from across the semiconductor value chain.
Its objectives include collaborating on common approaches to reducing GHG emissions, setting decarbonization targets to reach net zero by 2050, and reporting its progress toward these targets in relation to Scope 1, 2 and 3 emissions.
In May 2021 the International Energy Agency published its Flagship report, “Net Zero by 2050.” The report recognizes not all countries have the same start point and end date.
Scope 1, 2 and 3 targets
Scope 1 covers direct emissions; Scope 2 and Scope 3 are indirect.
Scope 1 refers to emissions owned, controlled or created by the company. This includes manufacturing and company-owned vehicles but can be counterbalanced by self-generated electricity.
Scope 2 emissions are indirect GHGs associated with the purchase of electricity, steam, heat or cooling. They physically occur at the facility where these utilities are consumed. These are indirect upstream (supplied) utilities.
Scope 3 is the indirect emissions from downstream activity. They are generated by assets not owned or controlled by the reporting organization. As a distributor for semiconductor manufacturers, Avnet’s activities as used by a reporting organization fall into Scope 3.
The scope of GHG emission reduction
The three scopes were developed by the GHG Protocol, backed by the World Resources Institute and the World Business Council for Sustainable Development. They have since been widely accepted as a fair way to calculate GHG emissions.
For semiconductor manufacturers, Scope 1 relates to the GHG emitted by the manufacturing process and ongoing efforts to abate emissions. Scope 2 is related to the size of the fab; a larger facility will use more power and resources. Depending on the size of the manufacturer’s fab and its position on its environmental journey, the balance between Scope 1 and Scope 2 can vary significantly.
Many manufacturers have announced plans to move to using renewable energy. Often those plans are to reach 100% renewable by or before 2050. RE100 is a renewable energy initiative with members from across all verticals.
However, moving to 100% renewable energy is only part of the ambition. Many companies also want to have a net-zero environmental impact, which will take longer and require much more than a switch to renewable energy sources.
The World Semiconductor Council is where semiconductor industry associations from around the globe come together. The WSC recently committed to achieving a PFC emissions reduction of 85% by 2030.
GHG abatement in semiconductor manufacturing
A lot of the effort will be focused on the activities in the sub-fab. This is the backroom to a fab, where machines work tirelessly to maintain the cleanroom air quality, but also remove and process the used resources, such as the chemicals used in semiconductor fabrication.
The abatement systems are designed to remove all harmful substances from the water used, for example. But achieving a sub-fab that is both carbon neutral and emission free is going to be difficult.
The main GHG contributor in the process is the perfluorinated compound gasses, or PFCs, used in the etching, cleaning and testing stages of wafer processing. In May 2021, the European Semiconductor Industry Association (EISA) announced a 42% absolute emission reduction in PFCs across the industry, from 2010 to 2020.
EISA explained this had been achieved by process optimization, increased efficiency, the use of alternative chemistries, and the installation of abatement equipment. The global effort to reduce PFC emissions is managed by the World Semiconductor Council.
Improving efficiency in chips and their fabrication
To continue the trend toward higher integration, semiconductor fabrication is in the stage of moving from FinFETs to nanosheets. Seen as an evolution of vertical transistor technology, nanosheets enable a continued scaling down in cell size. This will result in more transistors per chip and higher feature integration.
Moreover, some reports predict as much as an 85% reduction in active power when using nanosheets instead of FinFETs. That could impact the energy needed in applications where integrated circuits (ICs) are a main consumer. Mobile devices would operate longer on a single charge, and data centers could use significantly less energy both for running the processors and cooling them.
Manufacturing semiconductors has grown more complex as integration has increased. The number of process steps doubled from around 500 for the 90nm planar node to over 1,000 for 40nm FinFETs. Each process consumes energy, resources and results in GHGs.
The introduction of extreme ultraviolet (EUV) lithography reduced the number of process steps needed as scaling continued and, as a result, the complexity of manufacturing. But as the smaller nodes enable higher integration, chips are getting bigger and more featured. Complexity is still on the increase. The result: Associated emissions are expected to double between 2015 and 2025.
Couple these figures with an increased number of fabrication plants being built around the world, and it is clear to see why efforts to make the semiconductor supply chain more environmentally sustainable are also increasing.
According to a blog published by Cadence, nanosheet, FETs can break the current power density barrier of 100 W per square centimeter, established by FinFET technology.
Like semiconductors, AI will be transformative in every aspect of modern life. And like the integrated devices that will run the AI algorithms, most of it will be hidden from view.
Using AI sustainably
The success and demand for artificial intelligence (AI) has created a somewhat warped version of Moore’s Law. The compute power used to train some large language models is doubling every three months. This puts data centers at the forefront of energy use.
Putting AI at the network’s edge makes a lot of sense. It reduces energy consumption in data centers by moving the processing effort away from the core. Ultra-low power microprocessors and even microcontrollers are now capable of running inference engines locally in the endpoint.
With billions of endpoints forming the IoT, this may not be the solution it first seems. Researchers at Technological University Dublin highlight the relationship between energy consumed and the precision achieved. The more bits in a floating-point operation, the more energy consumed. The more data moved into and out of static and dynamic RAM, the more energy consumed.
They see the challenge being to reduce the precision, and therefore number of multiply/accumulate operations and memory accesses, without losing accuracy. One recommendation is to change the way we train AI models. Instead of training small models for a long time, we should train small models for less time, and then heavily compress, or prune, the results.
If AI is part of the problem, then it is also part of the solution. IBM believes it can use AI to drastically reduce the time and research effort needed to discover new materials that could be used in semiconductor fabrication.
This could result in a reduction in the use of PFCs, as the industry could substitute it for a chemical with lower GHG emissions.
ON MARKETWATCH: CEO Phil Gallagher discusses how data-driven supply chains can reduce carbon emissions and boost earnings.
WHAT COMES NEXT IS UP TO ALL OF US
OEMs dependent on the semiconductor industry have no direct influence over how it develops. The aggregated demand from the market has always directed innovation. Legislation and a willingness to address environmental concerns are perhaps now taking precedence.
Product manufacturers enabled by semiconductor technology can still act. Designing for reuse and recyclability is one option.
Modern electronic products are mostly connected, often software defined and typically updated over the air during their lifecycle. This is creating a platform for hardware that is more homogenous and more easily repurposed.
The trend toward “as a service” is also an opportunity to embrace sustainability. The hardware used to deliver the service may no longer be the value proposition. Designing hardware for a specified number of operational hours, after which they are retrieved and recycled, could present a more sustainable business model for as-a-service providers.
The industry is not currently aligned around this circular philosophy. Many may feel strongly that products need to have a long operational lifetime to offset their carbon footprint, and that may be true. With legislation regulating the disposal of electronic waste varying hugely around the world, the right way forward is unclear.
As a member of the Electronic Components Industry Association (ECIA) and participant in its Global Industry Practices Committee, Avnet is working with other members to address environmental issues in the industry.
Avnet is also a corporate member of the Semiconductor Industry Association (SIA). The SIA operates the Semiconductor PFAS Consortium, which is working hard to identify how per- and poly-fluoroalkyl (PFAS) substances are used, how their use can be reduced, and find potential alternatives. The PFAS group of chemicals are synthetically produced and used extensively, not only in semiconductor production. These chemicals do not break down naturally, and the impact of their production and wider use is an ongoing area of research.
What is clear is that electronics engineers are adept to change and are in a strong position to effect positive change for a more sustainable electronics industry.
What’s next in sustainable AI?
IBM is also looking at the next step in AI. It suggests moving from deep learning, which uses large amounts of labeled data, to foundation models, which use massive amounts of unlabeled data. Foundation models also use self-supervised learning.
Self-supervised learning could be six times faster and five times cheaper than supervised learning. And foundation models could reduce the time taken to get to a trained model by a factor of six.
Another innovation, which echoes the earlier observation about energy used, is the development of analog in-memory compute (AIMC). The approach to AIMC varies by manufacture, but typically uses electrically controlled variable resistors in memory cells to act as weights in a neural network. In this way, the entire memory array becomes a layer in a network. Operating in the analog domain avoids the latency and power needed to convert and move data to and from the digital domain. The technology is nascent but holds a lot of promise.
What’s next in sustainable IC fabrication?
Industry 4.0 doesn’t just benefit conventional manufacturers. Semiconductor fabrication also needs to be smarter. The complexity involved means the individual gains may be small but together will add up to high value.
Automation and predictive maintenance are two examples. Both will require maximizing the potential of the huge amounts of data produced by a sub-fab.
ABOUT THE AUTHOR
Editor-in-Chief and Senior Technology Writer, Avnet
Philip leads our FAE roundtable discussions and develops content covering the full range of technologies supported by Avnet.
Philip has more than 30 years of electronics industry experience, including working as a design engineer on mixed-signal embedded systems. He was also a technical journalist and editor covering the industry for several European technical magazines. He has worked for small, medium and large companies as well as startups, and is pleased to say he is constantly learning.
He holds a post-graduate diploma in advanced microelectronics.